Cluely Rival Sett Raises $27M for AI Bot Builders 2026

Sett, a Tel Aviv startup most business owners have never heard of, just closed a $27M Series A extension and opened its self-serve Sett AI creative agent to advertisers outside mobile gaming. That matters more than another foundation-model announcement, because Sett is not selling intelligence — it sells finished ad creatives that get tested, scored, and killed automatically. While the headline labs argue about benchmarks, a 40-person team generates thousands of video and static ad variants a week and lets the ad networks decide which ones survive. If you buy paid traffic for anything — a Shopify store, a SaaS trial, a local service business — the economics of creative production changed this week.

What’s new about the Sett AI creative agent

Sett has operated in mobile game user acquisition since 2023, where it built an agentic pipeline that ingests a game’s assets, generates ad concepts, renders them into playable and video creatives, ships them live to Meta, Google, TikTok, AppLovin, and Unity, then reads back performance data and iterates. The Series A extension brings total funding to roughly $42M and is led by existing investors doubling down on a company that already counts several top-grossing mobile publishers as customers. The Sett Series A funding round is notable less for its size than for what it funded: not more research, but more GPU capacity and more ad-network integrations.

The new part is the self-serve tier. Until this week, Sett ran an enterprise motion — you talked to a human, signed an annual contract, and got a managed creative team-in-a-box. The self-serve product lets a non-gaming advertiser connect an ad account, upload a brand kit and a product feed, and get a running creative testing loop without a sales call. That opens AI ad creative generation to e-commerce, apps, D2C, and lead-gen advertisers who were never going to buy a six-figure gaming tool.

The architecture is the interesting bit for anyone evaluating agentic marketing tools 2026. Sett does not use one model. It orchestrates a chain: a research agent that scrapes competitor creatives and app store listings, a concept agent that writes hooks and scripts, a production agent that calls video and image models plus a template renderer, a QA agent that rejects broken renders and off-brand output, and a media-buying agent that manages budget allocation across variants. Each stage is boring on its own. The compounding effect is that a business can test 200 creative concepts in the time a traditional agency ships four.

Why it matters

• Creative volume is the last real lever in paid ads. Meta and Google took targeting away from you and gave it to their algorithms. You still control the creative you feed them. Whoever produces more testable variants wins, and that is now a software cost, not a headcount cost.
• The $3,000/month freelance video editor is under direct pressure. Not eliminated — someone still has to define brand voice and approve output — but the “make me 30 versions of this ad with different hooks” job is gone. Reprice that budget toward strategy and offer testing.
• It proves the application layer is where the money is. Sett did not train a frontier model. It wrapped existing ones in a workflow that maps to a line item advertisers already pay for. That is the template for profitable AI businesses in 2026, and a small company can copy it in an adjacent vertical.
• AI UA creative testing changes your reporting cadence. If your agency reports monthly on six creatives, they are structurally slower than a competitor running automated daily iteration. Ask them what their weekly variant count is. The answer tells you whether to renew.
• Ad platforms will absorb some of this. Meta’s Advantage+ creative and Google’s asset generation already do a weaker version for free. Sett’s defensibility is cross-network testing and creative memory — knowing which hooks worked for your account, not just this campaign.
• Brand safety becomes an operations problem. When a machine ships 400 creatives a week, one bad claim or one off-brand render is legal exposure. You need an approval gate before spend, not after.

How to use the Sett AI creative agent today

You do not need Sett to capture most of this value. Here is how to stand up a creative testing loop this week, whether you sign up for a self-serve AI ad platform or build a lightweight version yourself.

1. Audit your current creative velocity first. Pull the last 90 days from your ad account and count distinct creatives, not campaigns. If the number is under 20, volume is your bottleneck and any automation will pay for itself.

   # Meta Marketing API — count distinct creatives and spend by ad, last 90 days
   curl -G "https://graph.facebook.com/v21.0/act_<YOUR_AD_ACCOUNT_ID>/insights" \
     -d "level=ad" \
     -d "fields=ad_id,ad_name,spend,impressions,actions,cost_per_action_type" \
     -d "date_preset=last_90d" \
     -d "limit=500" \
     -d "access_token=$META_ACCESS_TOKEN"

2. Write a creative brief the agent can consume. Agentic tools fail on vague brand input. Put your constraints in a structured file and reuse it for every generation run.

   brand:
     name: "Northgate Supply"
     voice: "direct, no hype, trades-focused"
     banned_claims: ["guaranteed", "#1", "risk-free", "lowest price"]
     must_include: ["free shipping over $75", "30-day returns"]
   audience:
     primary: "independent contractors, 28-55, US, mobile-first"
     pain_points: ["job site downtime", "parts that fail in cold weather"]
   offer:
     hook_angles: ["cost of downtime", "cold-weather durability", "next-day delivery"]
     cta: "Shop the winter kit"
   format:
     ratios: ["9:16", "1:1", "4:5"]
     video_length_s: [6, 15]
     first_frame_rule: "product visible within 0.5s, text hook under 6 words"

3. Generate hook variants in batches, not one at a time. This prompt structure is what the concept stage of a Sett-style pipeline does. Run it against whatever model you already pay for.

   You are a direct-response UA creative strategist.
   
   INPUT: the brand YAML below.
   TASK: produce 25 distinct ad concepts. Each concept must vary on ONE
   axis at a time so the test is readable: hook angle, opening visual,
   or CTA placement. Never vary two axes in the same concept.
   
   For each concept return JSON:
   {
     "id": "c01",
     "axis_varied": "hook_angle",
     "hook_text": "<=6 words, on-screen at 0.0s",
     "vo_script": "<=30 words",
     "opening_visual": "one sentence, shootable or renderable",
     "cta": "string",
     "predicted_failure_mode": "why this one might flop"
   }
   
   HARD RULES: no banned_claims. No emojis. No superlatives without
   a cited number. Output valid JSON array only.
   
   BRAND YAML:
   <paste the file from step 2>

4. Ship them into a structured test, not one blended campaign. The mistake that kills AI UA creative testing is dumping 25 variants into one ad set and letting the algorithm concentrate spend on the first lucky winner. Cap spend per variant until each clears a minimum impression floor.

   {
     "test_config": {
       "min_impressions_before_judgment": 3000,
       "max_spend_per_variant_usd": 40,
       "kill_rule": "cpa > 1.6 * account_median_cpa AND impressions >= 3000",
       "scale_rule": "cpa < 0.8 * account_median_cpa AND conversions >= 5",
       "refresh_cadence_days": 7,
       "winners_carried_forward": 3
     }
   }

5. Build a creative memory file. This is the durable asset — more valuable than any single winning ad. After each cycle, log what won and why, then feed it back into step 3 as context. Six cycles in, your prompts beat any agency’s starting point.

   echo "$(date +%F) | c14 | axis=hook_angle | angle=cost_of_downtime | CPA \$18.40 vs median \$31.10 | WINNER | note: problem-first hooks beat product-first 4:1 this quarter" \
     >> creative_memory.log

6. Put a human approval gate before spend. One person, five minutes, checking every batch against the banned-claims list and brand kit. This is non-negotiable at volume.

How it compares

Tool	Best for	Creative generation	Auto media buying	Cross-network testing	Entry point
Sett	Volume UA testing, now beyond gaming	Video, static, playables	Yes — full loop	Meta, Google, TikTok, AppLovin, Unity	Self-serve as of this week
Meta Advantage+ Creative	Advertisers already all-in on Meta	Variations on your uploads	Within Meta only	No	Free, inside Ads Manager
AppLovin SparkLabs	Mobile game publishers at scale	Video and playables	Within AppLovin	No	Managed, spend-gated
Creatopy / AdCreative.ai	SMB static and display output	Mostly static, template-driven	No	No	Low-cost monthly SaaS
Traditional agency	Brand campaigns, complex production	Human, high quality, low volume	Human-managed	Yes, manually	Retainer, typically $5k+/mo
DIY pipeline (steps above)	Owners who want to keep the IP	Whatever models you already pay for	No — you ship manually	Manual	Your existing LLM subscription

The honest read: if you spend under roughly $20k/month on paid ads, the DIY pipeline plus your existing tools gets you most of the way. Above that, the automated buying loop is where a tool like Sett earns its fee, because human reaction time becomes the bottleneck.

What’s next

Watch whether the self-serve tier retains non-gaming customers. Mobile gaming is an unusually forgiving market for automated creative: the product is visual, the conversion event is an install, and creative fatigue is brutal enough that volume genuinely wins. E-commerce and B2B are harder — longer consideration, more brand risk, messier attribution. If Sett’s non-gaming cohort churns at 90 days, that tells you mobile game ad automation was the moat, not the AI.

The second thing to watch is platform encroachment. Meta and Google both have every incentive to make creative generation free and native, because more creative variety means more auction liquidity for them. Sett’s answer has to be cross-network intelligence — the ability to say “this hook wins on TikTok and dies on Meta, here’s the budget shift” — which no single platform will ever build for you. That is a defensible position, but it requires the ad networks to keep granting API access on reasonable terms. They have revoked it before.

For business owners, the practical roadmap item is this: expect the “creative agent” category to consolidate fast over the next 12 months, with three or four winners and a graveyard of thin wrappers. Do not sign a long annual contract with anyone in this space yet. Do start building your creative memory log now, because that data is portable and it makes any of these tools work well for your specific account.

Frequently Asked Questions

Do I need to be a mobile game company to use Sett?

Not as of this week. The self-serve tier is open to general advertisers. The product was tuned on game UA data, so expect the video output to skew toward fast-cut, hook-forward formats. If your category needs slow, considered brand storytelling, this is the wrong tool.

What does the Sett AI creative agent cost?

Enterprise deals have historically been performance-linked, tied to a percentage of managed ad spend. Self-serve pricing is new and will likely shift — treat any number you see this month as provisional and negotiate on your actual spend level rather than accepting list pricing.

Will AI-generated ads get my account flagged?

Generated creative is not itself a policy violation on any major network, but volume raises your odds of tripping something. The real risks are unsubstantiated claims, generated faces that imply endorsement, and before/after imagery in restricted categories. The banned-claims list and human approval gate in step 6 above exist for this.

How much ad spend justifies an agentic marketing tool?

Roughly $20k/month is where the automated buying loop starts to beat a competent human checking dashboards twice a day. Below that, the DIY prompt pipeline captures most of the creative-volume benefit at near-zero marginal cost, and you keep the creative memory as your own asset.

Can I replace my agency with this?

You can replace the production half. You cannot replace offer strategy, positioning, or the judgment call about which of four business angles to test this quarter. Most owners who try full replacement end up with high-volume, well-produced ads for a weak offer — a faster way to lose money, not a cheaper one.

Why does the Series A extension matter if the product already existed?

Because it funded distribution rather than research. A $27M extension going into self-serve onboarding and ad-network integrations signals that the technical risk is considered solved and the remaining question is customer acquisition. For you as a buyer, that means the product is stable enough to evaluate on results rather than promises.